3. Windows Scheduling
• Priority Boost when unblocking
– Actual boost dependent on resource
• Disk (1), serial (2), keyboard (6), soundcard (8)…..
• Interactive, window event, semaphore (1 or 2)
– Boost decrements if quantum expires
• Anti-starvation hack
– If a ready process does not run for long time,
it gets 2 quanta at priority 15
3
5. Batch Algorithms
– Maximise throughput
• Throughput is measured in jobs per hour (or similar)
– Minimise turn-around time
• Turn-around time (Tr)
– difference between time of completion and time of submission
– Or waiting time (Tw) + execution time (Te)
– Maximise CPU utilisation
• Keep the CPU busy
• Not as good a metric as overall throughput
5
6. First-Come First-Served (FCFS)
• Algorithm
– Each job is placed in single queue, the first
job in the queue is selected, and allowed to
run as long as it wants.
– If the job blocks, the next job in the queue is
selected to run
– When a blocked jobs becomes ready, it is
placed at the end of the queue
6
7. Example
• 5 Jobs
J1
– Job 1 arrives slightly
before job 2, etc…
J2
– All are immediately
J3 runnable
– Execution times
J4 indicated by scale on
x-axis
J5
0 2 4 6 8 10 12 14 16 18 20
7
9. FCFS
• Pros
– Simple and easy to implement
• Cons
– I/O-bound jobs wait for CPU-bound jobs
⇒Favours CPU-bound processes
• Example:
– Assume 1 CPU-bound process that computes for 1 second and
blocks on a disk request. It arrives first.
– Assume an I/O bound process that simply issues a 1000
blocking disk requests (very little CPU time)
– FCFS, the I/O bound process can only issue a disk request per
second
» the I/O bound process take 1000 seconds to finish
– Another scheme, that preempts the CPU-bound process when
I/O-bound process are ready, could allow I/O-bound process to
finish in 1000* average disk access time.
9
10. Shortest Job First
• If we know (or can estimate) the execution
time a priori, we choose the shortest job
first.
• Another non-preemptive policy
10
11. Our Previous Example
• 5 Jobs
J1
– Job 1 arrives slightly
before job 2, etc…
J2
– All are immediately
J3 runnable
– Execution times
J4 indicated by scale on
x-axis
J5
0 2 4 6 8 10 12 14 16 18 20
11
13. Shortest Job First
• Con
– May starve long jobs
– Needs to predict job length
• Pro
– Minimises average turnaround time (if, and only if, all
jobs are available at the beginning)
– Example: Assume for processes with execution times
of a, b, c, d.
• a finishes at time a, b finishes at a + b, c at a + b + c, and so
on
• Average turn-around time is (4a + 3b + 2c + d)/4
• Since a contributes most to average turn-around time, it
should be the shortest job.
13
14. Shortest Remaining Time First
• A preemptive version of shortest job first
• When ever a new jobs arrive, choose the
one with the shortest remaining time first
– New short jobs get good service
14
15. Example
• 5 Jobs
J1
– Release and execution
times as shown
J2
J3
J4
J5
0 2 4 6 8 10 12 14 16 18 20
15
24. Three Level Scheduling
• Admission Scheduler
– Also called long-term scheduler
– Determines when jobs are admitted into the
system for processing
– Controls degree of multiprogramming
– More processes ⇒ less CPU available per
process
24
25. Three Level Scheduling
• CPU scheduler
– Also called short-term scheduler
– Invoked when ever a process blocks or is
released, clock interrupts (if preemptive
scheduling), I/O interrupts.
– Usually, this scheduler is what we are
referring to if we talk about a scheduler.
25
26. Three Level Scheduling
• Memory Scheduler
– Also called medium-term scheduler
– Adjusts the degree of multiprogramming via
suspending processes and swapping them
out
26
27. Some Issues with Priorities
• Require adaption over time to avoid starvation
(not considering hard real-time which relies on
strict priorities).
• Adaption is:
– usually ad-hoc,
• hence behaviour not thoroughly understood, and
unpredictable
– Gradual, hence unresponsive
• Difficult to guarantee a desired share of the CPU
• No way for applications to trade CPU time
27
28. Lottery Scheduling
• Each process is issued with “lottery
tickets” which represent the right to
use/consume a resource
– Example: CPU time
• Access to a resource is via “drawing” a
lottery winner.
– The more tickets a process possesses, the
higher chance the process has of winning.
28
29. Lottery Scheduling
• Advantages
– Simple to implement
– Highly responsive
• can reallocate tickets held for immediate effect
– Tickets can be traded to implement individual
scheduling policy between co-operating
threads
– Starvation free
• A process holding a ticket will eventually be
scheduled.
29
30. Example Lottery Scheduling
• Four process running concurrently
– Process A: 15% CPU
– Process B: 25% CPU
– Process C: 5% CPU
– Process D: 55% CPU
• How many tickets should be issued to
each?
30
33. Fair-Share Scheduling
• So far we have treated processes as individuals
• Assume two users
– One user has 1 process
– Second user has 9 processes
• The second user gets 90% of the CPU
• Some schedulers consider the owner of the process in
determining which process to schedule
– E.g., for the above example we could schedule the first user’s
process 9 times more often than the second user’s processes
• Many possibilities exist to determine a fair schedule
– E.g. Appropriate allocation of tickets in lottery scheduler
33
